arXiv subjects

Letian Yang

Publications and source records attributed to Letian Yang.

4 recordsLinked to original sources

arXiv2026

ROAD: Adaptive Data Mixing for Offline-to-Online Reinforcement Learning via Bi-Level Optimization

Offline-to-online reinforcement learning harnesses the stability of offline pretraining and the flexibility of online fine-tuning. A key challenge lies in the non-stationary distribution shift between offline datasets and the evolving online policy. Common approaches often rely on static mixing ratios or heuristic-based replay strategies, which lack adaptability to different environments and varying training dynamics, resulting in suboptimal tradeoff between stability and asymptotic performance. In this work, we propose Reinforcement Learning with Optimized Adaptive Data-mixing (ROAD), a dynamic plug-and-play framework that automates the data replay process. We identify a fundamental objective misalignment in existing approaches. To tackle this, we formulate the data selection problem as a bi-level optimization process, interpreting the data mixing strategy as a meta-decision governing the policy performance (outer-level) during online fine-tuning, while the conventional Q-learning updates operate at the inner level. To make it tractable, we propose a practical algorithm using a multi-armed bandit mechanism. This is guided by a surrogate objective approximating the bi-level gradient, which simultaneously maintains offline priors and prevents value overestimation. Our empirical results demonstrate that this approach consistently outperforms existing data replay methods across various datasets, eliminating the need for manual, context-specific adjustments while achieving superior stability and asymptotic performance.

cs.LG↗

arXiv2026

MDAgent: A Multi-Agent Framework for End-to-End Molecular Dynamics Research

Molecular dynamics (MD) simulation is a powerful tool for studying biomolecular structural changes, molecular recognition, transmembrane transport, and functional mechanisms. However, its practical bottleneck lies not only in software operation or parameter setup, but in translating experimental questions into executable, interpretable, and reviewable computational workflows. Here, we present MDAgent, a multi-agent system for end-to-end molecular dynamics research. The system integrates problem understanding, literature-guided strategy design, simulation execution, trajectory analysis, mechanistic interpretation, and quality supervision into a unified workflow, enabling agents not only to run simulations but also to generate research-oriented computational plans and analytical reports. We further introduce a case-based learning mechanism based on Skill and Memory, which stores reusable knowledge from prior tasks, including parameter choices, operational rules, analytical logic, and problem-solving pathways, thereby supporting cross-task transfer without retraining the underlying model. Across multiple representative molecular simulation tasks, MDAgent achieved stable end-to-end performance with improved strategic adaptability, interpretability, and generalization. In an independent complex task involving conformational transitions of TMEM16F and XKR8, the system successfully completed system design, simulation, and mechanistic analysis for large membrane proteins. These results show that combining multi-agent collaboration with case-based learning can transform MD agents from workflow automation tools into scientific question-oriented computational research systems, providing a scalable framework for AI-driven automated research.

q-bio.QM↗

arXiv2026

Transferable Expertise for Autonomous Agents via Real-World Case-Based Learning

LLM-based autonomous agents perform well on general reasoning tasks but still struggle to reliably use task structure, key constraints, and prior experience in complex real-world settings. We propose a case-based learning framework that converts experience from past tasks into reusable knowledge assets, allowing agents to transfer prior case experience to new tasks and perform more structured analysis. Unlike methods based mainly on pretrained knowledge or static prompts, our framework emphasizes extracting and reusing task-relevant knowledge, analytical prompts, and operational skills from real cases. We evaluate the method on a unified benchmark of six complex task categories and compare it with Zero-Shot, Few-Shot, Checklist Prompt, and Rule Memory baselines. Results show that our method achieves consistently strong performance across all tasks and matches or outperforms the best baseline in every case, with especially clear gains on more complex tasks. Further analysis shows that the advantage of case-based learning increases with task complexity, and that practical knowledge acquired by one agent can be reused by others. These findings suggest that case-based learning offers a promising path for building professional agents for real-world work.

cs.AI↗

arXiv2024

Algebraic and Statistical Properties of the Partially Regularized Ordinary Least Squares Interpolator

Modern deep learning has revealed a surprising statistical phenomenon known as benign overfitting, with high-dimensional linear regression being a prominent example. This paper contributes to ongoing research on the ordinary least squares (OLS) interpolator, focusing on the partial regression setting, where only a subset of coefficients is implicitly regularized. On the algebraic front, we extend Cochran's formula and the leave-one-out residual formula for the partial regularization framework. On the stochastic front, we leverage our algebraic results to design several homoskedastic variance estimators under the Gauss-Markov model. These estimators serve as a basis for conducting statistical inference, albeit with slight conservatism in their performance. Through simulations, we study the finite-sample properties of these variance estimators across various generative models.

math.ST↗